Time Series Forecasting

Store-Sales Time Series Forcasting from Kaggle.

Introduction

This study will be forcusing on time-series forecasting for store sales. The data is extracted from Kaggle. The data is provided by an Ecuador company known as Corporación Favorita.

Methodology

I will be exploring the modeltime library to conduct nested forecasting for the dataset provided.

Dataset

There are a total of 54 stores and 33 product families in the data. The time series starts from 01 Jan 2013 and finishes in 31 Aug 2017. The data is splitted to train and test data, and the dates in the test data are 15 days after the last date in the training data.

The dataset consist of 6 different worksheets. The description of each worksheets is as such:

• Train.csv contains time series of the stores and the product families combination.

  • The store_nbr column identifies the store at which the products are sold.
  • The family column identifies the type of product sold.
  • The sales column gives the total sales for a product family at a particular store at a given date. Fractional values are possible since products can be sold in fractional units (1.5 kg of cheese, for instance, as opposed to 1 bag of chips).
  • The onpromotion column gives the total number of items in a product family that were being promoted at a store at a given date.

• test.csv contains similar features as the training data.

  • The dates in the test data are for the 15 days after the last date in the training data.

• Stores.csv contains the metadata of all the stores participating in this analysis.

  • The store_nbr column identifies the store at which the products are sold.
  • The city column identifies the city the store is located in.
  • The state column identifies the state the store is located in.
  • The type column identifies the type of the store.
  • The cluster column highlights the cluster the store belongs to.

• Transactions.csv contains the number of transactions recorded by stores throughout the period of analysis.

• Oil.csv contains the daily oil price. Includes values during both the train and test data time frames.

• holidays_events.csv contains all the different holidays and event that happen throughout the period of analysis.

  • The transferred column indicate if the holiday was celebrated on the actual day or was transferred to another day.

In addition to the data provided through the data set, there are two pointers to be aware of:

• Wages in the public sector are paid every two weeks on the 15 th and on the last day of the month. Supermarket sales could be affected by this.

• A magnitude 7.8 earthquake struck Ecuador on April 16, 2016. People rallied in relief efforts donating water and other first need products which greatly affected supermarket sales for several weeks after the earthquake

Libraries

These are the libraries used for the analysis.

pacman::p_load(tidyverse, tidymodels, 
               timetk, modeltime,ggstatsplot,lubridate, trelliscopejs, seasonal,
               tsibble, feasts, fable, forecast,psych)

oil <- read_csv ("rawdata/oil.csv")
holiday <- read_csv ("rawdata/holidays_events.csv")
test <- read_csv ("rawdata/test.csv")
train <- read_csv ("rawdata/train.csv")
stores <- read_csv ("rawdata/stores.csv")
transacation <- read_csv ("rawdata/transactions.csv")

describe(oil)

           vars    n  mean    sd median trimmed  mad   min    max
date          1 1218   NaN    NA     NA     NaN   NA   Inf   -Inf
dcoilwtico    2 1175 67.71 25.63  53.19   66.96 18.7 26.19 110.62
           range skew kurtosis   se
date        -Inf   NA       NA   NA
dcoilwtico 84.43 0.32    -1.61 0.75

oilplot <- ggplot(oil,aes(x = date,y=dcoilwtico)) +
  geom_line(colour = "#468499") +
  ylim(25,115) +
  theme_classic() +
  labs(y = "Oil Price", x = "Date", title = "Daily Oil Price from 2013 - 2017", subtitle = "Without Linear Interplolation") +
  theme(axis.title.y = element_text(angle=0))

oilplot 

ts_oil <- oil %>%
  as_tsibble(index = `date`)

ts_oil$dcoilwtico <- (na.interp(ts_oil$dcoilwtico))


oilplot <- ggplot(ts_oil,aes(x = date,y=dcoilwtico)) +
  geom_line(colour = "#D2288A") +
  ylim(25,115) +
  theme_classic() +
  labs(y = "Oil Price", x = "Date", title = "Daily Oil Price from 2013 - 2017", subtitle = "With Linear Interplolation") +
  theme(axis.title.y = element_text(angle=0))

oilplot 

We assume that the oil prices does not change during the weekends or holidays. Therefore, we use the fill function to fill up the weekends NA value based on the previous oil price values.

ts_oil_fill <- ts_oil %>%
  complete(date = seq.Date(min(date), max(date), by = "day" )) %>%
  fill (dcoilwtico)


ts_oilfill <- ggplot(ts_oil_fill,aes(x = date,y=dcoilwtico)) +
  geom_line(colour = "#D2288A") +
  ylim(25,115) +
  theme_classic() +
  labs(y = "Oil Price", x = "Date", title = "Daily Oil Price from 2013 - 2017", subtitle = "With Linear Interplolation and fill") +
  theme(axis.title.y = element_text(angle=0))

ts_oilfill

Relationship between Sales and Oil Prices

train_correlation <- train %>%
  group_by (date) %>%
  summarise (total_sales = sum(sales)) %>%
  ungroup()

train_oil <- train_correlation %>%
  left_join (ts_oil_fill, by = "date")

ggscatterstats(
  data  = train_oil,
  x     = total_sales,
  y     = dcoilwtico,
  xlab  = "Total Sales",
  ylab  = "Daily Oil Price",
  title = "Checking for Correlation between Sales and Oil prices ",
  type = "np"
)

Family Product against oil prices

In this analysis, we will analyse if the sales of every product family is affected by oil prices.

train_productfamily <- train %>%
  group_by (date,family) %>%
  summarise (total_sales = sum(sales)) %>%
  ungroup()

train_PF <- train_productfamily %>%
  left_join (ts_oil_fill, by = "date")

grouped_ggscatterstats(
  data  = train_PF,
  x     = total_sales,
  y     = dcoilwtico,
  grouping.var = family,
  xlab  = "Total Sales",
  ylab  = "Daily Oil Price",
  type = "np",
  plotgrid.args    = list(nrow = 8)
)